Repeatability splitter

Press three or more pellets from one ground powder and measure each of them two or more times, the same number of times each. Enter one pellet per line. Measure them in a mixed order (1, 2, 3, then 1, 2, 3 again) under the same conditions. The tool splits the scatter into the part within pellets and the extra part between them, compares the within-pellet part with counting statistics if you give the raw counts, and checks the total against your own target.

Classical one-way analysis of variance for a balanced design: s_meas from the scatter of repeats on each pellet, s_prep from the scatter of the pellet means beyond what s_meas explains, and s_total² = s_prep² + s_meas². It is not the robust analysis of variance of the sampling guides, so one outlying value pulls it, and with only a few pellets every estimate is uncertain. The split assumes a balanced design and independent repeats; it locates the scatter under the design used and does not name its cause. All pellets come from one ground powder, so grinding is not in the estimate, and if each pellet is measured in one block, instrument drift shows up between pellets. Counting statistics: a relative standard deviation of 1/√N for N raw counts of one line; it does not apply directly to a concentration calculated from several lines with corrections. The values shown when the tool opens are an example only. Whether the scatter is small enough for your work is your target, not the tool’s.